Problem

FallbackUseLock.incrementUserCount() in NdbUseLock.swift acquires ndbUserCountLock, then blocks on ndbAccessSemaphore.wait(). If another thread holds the semaphore and needs the lock, deadlock occurs.

Location

nostrdb/NdbUseLock.swift lines 148-156

private func incrementUserCount(maxTimeout: DispatchTimeInterval = .seconds(2)) throws {
    ndbUserCountLock.lock()
    defer { ndbUserCountLock.unlock() }
    
    // Signal that ndb cannot close while we have at least one user using ndb
    if ndbUserCount == 0 {
        try ndbAccessSemaphore.waitOrThrow(timeout: .now() + maxTimeout)  // BLOCKS WHILE HOLDING LOCK
    }
    ndbUserCount += 1
}

Root Cause

Classic lock ordering violation: Thread A holds lock, waits for semaphore. Thread B holds semaphore, needs lock. Deadlock.

Impact

• App freeze on iOS < 18 (FallbackUseLock is used)
• Timeout after 2 seconds but thread is blocked during that time

Suggested Fix

Release lock before waiting on semaphore, or refactor to avoid nested blocking primitives.

Related Issues

• Ndb ingester queue pthread_mutex_lock crash related to switching in/out of the app #2472 (Ndb ingester pthread_mutex_lock crash)
• Ndbtxn transactions shouldn't be held open across async boundaries #3335 (NdbTxn async boundaries)

Test Plan

• Add concurrent access test with ThreadSanitizer
• Test database close during active reads
• Verify no deadlock under contention

Changelog-Fixed: Fix potential deadlock in FallbackUseLock